FTC's 'Suspicion' of AI Antitrust Exemptions Re-Rates the $500B Nvidia-OpenAI Syndicate — and Forces the 2027 AI Buildout to Find Non-Circular Money

At a Georgetown University event on Sep 15, 2026, FTC Chairman Andrew Ferguson said \"everyone should be deeply suspicious\" about AI companies that \"simultaneously\" ask Washington for new regulations and for antitrust exemptions, because such carve-outs ask for \"barriers to entry that will insulate their incumbency from challenge.\" The comments were directed at Anthropic CEO Dario Amodei's recent proposal for a frontier-pacing safe harbor — a bill introduced Jul 23, 2026 by Sens. Adam Schiff (D-CA) and Jim Banks (R-IN) called the Collaboration on Adversarial Threats and Security Risks Act (S.5105). Reuters described Ferguson's remarks as the first public indication of how the Trump administration views that bid.

The pushback is not symmetric. OpenAI lobbyist Chris Lehane told Bloomberg that OpenAI has been working with Anthropic and Alphabet on safety and \"did not see the need for an antitrust waiver,\" while Cohere CEO Aidan Gomez criticized the proposal as incumbent protection. Ferguson's framing — that exemption-seekers are also the regulation-seekers — therefore lands hardest on Anthropic, the company with the largest concentrated compute purchase order book (more than $100B committed to AWS over a decade, per Anthropic's Apr 20, 2026 statement, plus a $65B Series H raise on May 28, 2026 at a $965B valuation).

On Aug 10, 2026, Nvidia announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to \"mobilize over $500 billion of third-party capital\" for AI compute infrastructure. Five days later, on Aug 17, 2026, Nvidia separately committed up to $105 billion in credit and compute for OpenAI's 10-gigawatt, $500 billion PORTS-Pike campus in Pike County, Ohio, to be built by SB Energy (a SoftBank subsidiary) and exclusively host Nvidia AI compute. Read individually, each is a financing deal; read together, they form the largest concentrated example of \"circular financing\" in any U.S. industry — and the structure Ferguson, the DOJ and the courts are now circling.

The Aug 31, 2026 \"three weeks in\" read on the syndicate is that none of the $500B is yet committed — it is a memorandum of understanding for capital that asset managers must individually originate, underwrite and close. The first tranche of named, dated projects is the variable that converts the headline into deployed risk. Ferguson's Sep 15 framing turns that variable into a policy one: any deal that resembles Nvidia lending to a customer that uses the proceeds to buy Nvidia chips is now a public target, not just a footnote in short-seller reports.

On Aug 19, 2026, at the Technology Policy Institute's Aspen Forum, Ferguson said the biggest AI antitrust risks are upstream — in chips, power, and the construction needed to build AI data centers — not among large language model developers, which he called healthy because Anthropic, OpenAI, Google and Meta release near-weekly models. The Sep 15 statement extends that logic: if upstream concentration is the problem, then the Nvidia-TSMC-SK Hynix-Micron chain is exactly where competition policy is now aimed. The same logic also re-rates the $500B Wall Street vehicle — its job is to finance upstream infrastructure, which is now the locus of regulatory concern.

On Sep 9, 2026, the Justice Department opened an investigation into whether Nvidia structured its reported $20 billion non-exclusive licensing deal with Groq to skirt antitrust review of an outright acquisition, the New York Times reported (Sep 9) and Bloomberg confirmed (Sep 10). That action came two weeks after PYMNTS reported (Aug 30) that Nvidia had suspended parts of its AI Compute Partnership Program less than two months after launch, with employees having warned that revenue-share terms of up to 50% above a threshold and GPU-lease backstops could \"invite antitrust scrutiny.\" The pattern is consistent: the regulator is now treating Nvidia's customer financing as a structural antitrust input, not a marketing channel.

Tracing where the $500B syndicate, the $105B Ohio guarantee, and the upstream capex actually land produces a map with very different exposures on each side.

The cleanest call is upstream. TSMC trades at ~31x trailing earnings with $444B in FY2026 revenue and a $2.17T market cap; Micron trades at ~21x trailing earnings on a $90B revenue base and a $1.04T market cap, with HBM mix still rising; SK Hynix trades at ~7.4x trailing earnings with ~85% gross margins — the cheapest memory exposure in the supply chain, with HBM concentration as the explicit driver. None of these three is exposed to the FTC's financing-structure target; all three benefit if policy focuses on upstream capacity rather than capping chip purchases.

The hyperscaler pair is the second cleanest call. Oracle's Q1 FY2027 (Aug 31, 2026) printed $19.3B in revenue, $1.58 in GAAP EPS, and a $417B market cap at ~23x trailing earnings — the Stargate and OpenAI Ohio commitments make it the most direct replacement for excluded self-coordination. Microsoft is the broadest hedge at $3.72T market cap and ~28x trailing earnings, with $331.8B in TTM revenue.

The riskiest pair is the AI cloud specialists. CoreWeave's Q2 2026 revenue of $2.575B (+112% YoY) was reported on Aug 12, 2026, but the company carries $35.6B of debt and a -25.4% net margin — a structure that depends on continued access to Nvidia-supported financing. Nebius' Q2 2026 revenue of $582.3M (+454% YoY) was reported the same day, with 98% of revenue from AI cloud and a much smaller balance sheet; the model is more durable but is still circular-financing dependent, and Nvidia is a named equity holder (Mar 13, 2026, $2B investment).

The $500B Nvidia-led AI financing structure is no longer just a financing story — it is the regulatory event, and the next 90 days will determine whether the 2027 buildout gets re-financed or rationed.

For investors, three things follow. First, watch the first closed project from the syndicate — not the headline target — because that is what reveals whether the structure has been regulatorily acceptable. Second, the upstream-versus-downstream split is now policy-blessed: Ferguson himself drew the line, which makes TSMC, SK Hynix, Micron, Oracle and Microsoft the structural winners regardless of how the financing case resolves. Third, CoreWeave is the canary — its $35.6B debt load and Nvidia-financing dependency mean its next quarterly report is the first public test of whether regulator-driven rationing has begun.